基于ViewPoint的软件构架描述

软件构架作为系统开发的蓝图,已成为软件工程领域的热点。在软件开发实践中,许多项目更为关注的是软件构架的设计,而忽略了软件构架的描述。事实上软件构架的详细精确描述是成功开发软件的根本保证。文中以UML为构架描述语言,引用IEEE 1471标准中Viewpoint的概念,提出了基于3大Viewpoint的软件构架描述方法。基于Viewpoint的软件构架描述方法有助于开发人员在大型项目开发中从更高层次把握系统构架,从而保证软件开发过程的顺利进行。     

一种基于3维全景图像技术的深度测量方法

3维全景图像技术是一种能够记录和显示全真3维场景的图像技术。该技术采用微透镜阵列记录空间场景，空间任意一点的深度信息只需通过一次成像即可直接获得。本文研究采用全景图像技术直接获取物体空间信息的方法。此方法首先从全景图像中抽提视图。视图是通过抽提全景图像中对应于每个微透镜下同一局部位置的点人工合成的。每幅视图包含了全景图像中对原来的物空间场景按照某一特定方向的平行投影记录信息。接下来通过分析全景图像的光学成像过程。推导了用来描述物体深度信息和其在对应的视图间的视差关系的深度方程。从而得出空间任一点的深度可以通过其在对应视图间的视差来求得。最后，通过运用全景图像测量火柴盒的厚度的实例，验证了这一方法的可行性。其结果一方面可用于全景图像的数据处理本身，另一方面可望为开发新型的深度测量工具提供理论依据。     

一种具有自终止特性的视点规划方法

提出了一种具有自终止特性的对未知物体模型自动3维测量的视点规划方法。首先在寻找最优视点阶段，利用群矢量链方法获得视觉传感器在下一个视点的观测方向，而视觉传感器在空间的确切位置参数是通过计算矢量场内的边界积分获得，其中获得最大边界积分的空间位置被定为下一个最优视点。同时为了保证对未知物体模型自动3维测量过程具有自终止特性，根据高斯定理，提出由测量所得的点云直接计算出物体体积的方法，并将连续两个视点下物体体积的变化量作为终止测量的依据来判断规划过程是否需要终止或继续。实验结果表明，该具有自终止特性的视点规划方法是可行而且有效的。     

谈“建机适用工程学”

本文提出“适用工程学”的概念并将“适用工程学”理论应用到建筑机械的维修使用和运输中去。     

固定视点体视图最大图幅分析及其辅助计算作图法研究

从分析体视视野着手，研究固定视点体视图的最大图幅。同时，依据体视投影原理，推导出一系列计算公式，并列出图表，以实现大幅面体视图的直接绘制。①     

基于局部特征与视点感知的车辆重识别算法

在车辆重识别任务中,车辆视角的多变性会影响算法的准确性。为了解决视角多变对重识别准确性的影响,本文提出了一种基于局部特征与视点感知的车辆重识别方法。首先,使用语义分割算法将车辆解构为正面、背面、侧面、顶部4个部分,以提高车辆的细粒度表征。通过设计一种车辆视点感知网络,来输出视点的预测概率信息,据此概率信息动态平滑地呈现车辆视点感知效果。利用视点感知效果,为车辆每个局部区域赋予不同的权重,达到缩短类内距离,扩大类间差距,减少视角变化对车辆重识别的影响。利用公开数据集进行实验,其中VeRi776数据集的mAP可达到80.9%。结果表明,本方法可有效提高车辆重识别精度。结合消融实验证明了视点感知的平滑表示对多视角下车辆重识别的有效性。     

Viewpoint and the recognition of people from their movements.

Observers can recognize other people from their movements. What is interesting is that observers are best able to recognize their own movements. Enhanced visual sensitivity to self-generated movement may reflect the contribution of motor planning processes to the visual analysis of human action. An alternative view is that enhanced visual sensitivity to self-motion results from extensive experience seeing one's own limbs move. To investigate this alternative explanation, participants viewed point-light actors from first-person egocentric and third-person allocentric viewpoints. Although observers routinely see their own actions from the first-person view, participants were unable to identify egocentric views of their own actions. Conversely, with little real-world experience seeing themselves from third-person views, participants readily identified their own actions from allocentric views. When viewing allocentric displays, participants accurately identified both front and rear views of their own actions. Because people have little experience observing themselves from behind or from third-person views, these findings suggest that visual learning cannot account for enhanced visual sensitivity to self-generated action. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Canonical views of dynamic scenes.

The visual recognition of dynamic scenes was examined. The authors hypothesized that the notion of canonical views, which has received strong empirical support for static objects, also holds for dynamic scenes. In Experiment 1, viewpoints orthogonal to the main axis of movement in the scene were preferred over other viewpoints, whereas viewpoints in line with the main axis were least preferred. Experiment 2 provided no empirical evidence for a recognition advantage of canonical viewpoints when presented during the initial learning phase, but Experiment 3 showed a cognitive advantage for canonical viewpoints if they were presented as test stimuli during the recognition test. Overall, the findings suggest that on a phenomenological level, viewers are consciously aware of such viewpoints, and, on a cognitive level, viewers benefit from canonical viewpoints in terms of recognition accuracy. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

参数曲面消隐新算法

提出一种基于线段被三角形遮挡的消隐新算法，并利用此法推出了遮挡情况随视点变化的数学关系式，进而给出一种适合于动态模拟的消隐算法。     

基于摄像机纵向运动的序列图像的实时漫游

对于摄像机纵向运动的序列图像，提出了一种基于极线几何约束关系的当前视点目标图像生成算法：1）利用基于傅立叶变换的方法，得到远视点源图像中和近视点源图像无对应点的区域，并用一种背景色将其填充，得到填充后的远视点源图像；2）利用极线的整体匹配性质在两幅源图像中确定遍历整个源图像的对应源极线；3）将所有对应源极线按灰度进行分段；4）用动态规划匹配法确定对应源极线上段与段之间的匹配关系；5）通过两条对应源极线插值合成一条目标极线，生成当前视点的目标图像；6）对上一步得到的目标图像的背景色区域加以处理，得到最终的当前视点目标图像．提出了一种链表数据结构存储每相邻两幅源图像之间的预处理信息．漫游时实时读取．开发了一个图像漫游器，使用户能在其中实现不同视点的实时漫游．